TabAttackBench
收藏资源简介:
TabAttackBench是一个用于评估对抗攻击在表格数据上的有效性和不可感知性的基准数据集。该数据集由11个表格数据集组成,包括混合数据和仅数值数据集,旨在解决当前研究在表格数据对抗攻击中忽视不可感知性的问题。TabAttackBench的创建过程涉及评估五个对抗攻击算法在四个模型上的表现,通过分析攻击的成功率和不可感知性,为设计更有效和不可感知的对抗攻击算法提供有价值的信息。该数据集的应用领域主要涉及提高机器学习模型在表格数据上的鲁棒性和安全性,以应对对抗攻击带来的威胁。
TabAttackBench is a benchmark dataset for evaluating the effectiveness and imperceptibility of adversarial attacks on tabular data. It consists of 11 tabular datasets, including mixed-type and numeric-only datasets, aiming to address the issue that current research on tabular data adversarial attacks neglects the requirement of imperceptibility. The development of TabAttackBench involves evaluating five adversarial attack algorithms across four machine learning models, and provides valuable insights for designing more effective and imperceptible adversarial attack algorithms by analyzing attack success rates and imperceptibility levels. The main application scope of this dataset is to improve the robustness and security of machine learning models on tabular data to counter threats posed by adversarial attacks.

- 1TabAttackBench: A Benchmark for Adversarial Attacks on Tabular Data昆士兰科技大学信息系统学院, 昆士兰科技大学数据科学中心, 清华大学软件学院, 里斯本新里斯本大学NOVA信息管理学校, 悉尼科技大学数据科学研究所, 里斯本大学高级技术学院 · 2025年



